About this site, and what happened in the last 6 weeks since I discovered Claude.

TLDR: I discovered Claude and started to make projects with it. Some of the results weren’t great so I decided to teach it how to build software like I do with the teams I’ve worked with during my career, and give it tools so it can really test what it makes.

---

This website’s been something I’ve been thinking about for a little while. As you can probably tell by the main page of the site, I’m a software developer that’s been around the block a few times. I’ve been thinking about AI and experimenting with it a bit over the last couple of years, like a lot of coders have. My use, up until recently, was generating a small snippet here and there.

About six weeks ago I decided to give Claude a go, and that really changed a lot of things for me. It’s kinda delivered on the promise I was hoping AI software development would deliver. I started off with a really small project: just an app that was a central location to put a load of URLs for systems I was running on my home server. Then I discovered it’s quite a competent system so I decided I was going to try to build something more ambitious.

So I moved on to another thing I’d wanted to do for a while: my own version of Google Keep. I think Keep is awesome, but there are a few things about it I wanted customised to my own way of working. Semantic search. Being able to have AI, or things from the command line, push stuff into my notes from different places. I called it dKeep. There’s a version of it linked from this site, and it’ll be on GitHub too, assuming I’ve actually got that done by the time this goes out.

Whilst working on dKeep, I realised something interesting. AI is essentially guessing the next most likely word to fit the context, and when it writes code it’s doing the same thing, so it makes mistakes. I also realised it doesn’t really know good software development practice. So I thought, I’m going to teach it how to do that, just like I did with the teams that have worked with me. I started making a load of tools so it could test the software it was making.

Things got a bit out of hand, you could say. In about six weeks I built around 46 different projects. Obviously not all of them useful, but anything from a terminal I’d always wanted that runs on my phone and my computers, to a media player that runs on my computers and my phone and synchronises all the instances together. Quite recently I’ve been experimenting with whether you can develop MMO games with AI.

Over time I’ve really realised LLMs don’t “see” very well, even the multimodal ones. Claude get’s a reduced resolution image and can’t measure coords of things in images well. You’ll notice that the VNCPool and ADBPool applications have tools for zooming in, using OCR, and finding bits of things on the page that they can click, etc.

Other things I’ve developed are apps for getting dev app updates onto my mobile phone, and app updates pushed out to the different computers I use. There’s an app called Pusher for that. Other things are really simple: I made a very basic shopping list app called listShop. I was using a commercial app before, but I just wanted a simple version that worked the way I do, without all the adverts and the other nonsense.

Then I found myself wanting to talk to one Claude and have it orchestrate others, and that became the control app. Then, when some Claude instances were finding issues with tools made by other Claude instances, I made doable: an app that basically has todo lists for different bots and for me, so you can kick off lots of different bug fixes at the same time, with worktrees and all that.

It got to the point where I thought, actually, some of this might be useful to other people. It’s a cool system to have and be able to share. I’ve really enjoyed doing this stuff.

I’m going to talk on other blog posts about how I’ve found the whole process, and what software development is like for me using this team rather than working with human engineers. How I’ve changed my development process and my thought processes around software development. I know this sort of Ai stuff makes a lot of software developers uncomfortable, and I can understand why, but I can talk about my process and how I’ve been handling it. What I think about the code that comes out, how much I get involved in it. Surprisingly not as much as I thought I would, to be honest.

If anyone at Anthropic, or one of the other Ai companies out there, sees my projects finds them interesting, and wants to offer me a job, I’m more than happy to consider it. I’m in Zurich and I’m staying here, but if you’re good with me working remotely or have offices in Zurich, I’m more than interested. Get in touch.

Anyway, enjoy the site, enjoy the tools, and I’ll carry on updating things as I go along. Feel free to get in touch if there’s something here that takes your fancy or makes you interested to communicate. Catch you later.

Darren